Het samenvatten van rassenproeven en het toepassen van vruchtbaarheidscorrecties met niet-orthogonale methoden

Although a formula should be adequate, inadequate formulae had often to be used through lack of knowledge. With adequate formulae weighting of data was dependent on error expectation. With inadequate formulae weighting should be focused on representativeness. Some other pitfalls were discussed.
Correlation analysis was contrasted with regression analysis. These two models had in common that items of a universe were stochastically dependent on a functional relationship. Essential for correlation was that both correlated variables obeyed a probability distribution; for regression analysis this did not need to be true. In correlation the principal axis was a sound base for understanding relationships, while the regression line was a sound base for predictions. The contrast between understanding and prediction was discussed at large.

Special problems of pooling experiments were discussed. In principle a transformation of yield data was advisable if the interaction variety x experiment could be thus reduced. Yet the transformed yields remained inadequate. So weighting, if used, had to improve representativeness. A missing-plot technique for pooling was given.

A separate part deals with the reduction of residual error by drawing freehand fertility maps. This may be much more effective than the corrections provided by the design. A criterion was developed for avoiding undue detail in free-hand curves.

/HTML>